Launching today
DejaLoop
A live brake for AI coding agents stuck in loops
1 follower
A live brake for AI coding agents stuck in loops
1 follower
AI coding agents get stuck — re-reading the same file, retrying the same failing command, burning tokens while cost tools only show the damage after the session ends. DejaLoop is an open-source Claude Code plugin that intervenes live: deterministic checks fingerprint every tool call, a savings counter shows what stopped loops saved, and a semantic judge catches smart loops where every call differs but nothing moves. Fail-open, shadow mode by default. Free, MIT-licensed, no telemetry.
Launch Team
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
Hey Product Hunt 👋 I'm the maker of DejaLoop.
I built this after watching one too many Claude Code sessions quietly loop — same file read over and over, same test failing six different ways — while the token meter ran. The dashboards I used told me the cost the next day. I wanted a brake pedal instead of a report.
Two things I obsessed over: it can never block a healthy session by accident (fail-open, shadow mode first), and it costs nothing to run — the built-in judge is local, and the optional Jev judge runs about $0.0002 a session on your own key.
It's open source and I'd genuinely love your feedback — especially the hard kind. What loop patterns has your agent fallen into? What would a tool have to prove before you'd let it interrupt your agent?
I'll be here all day answering questions.